BACKGROUND

There are three major approaches that have been used to study and analyse the behaviour of FX market in the literature: behavioural finance, empirical models, and Agent-Based Modelling (ABM).

The Behavioural Finance approach is a psychology-based theory used to explain the be­haviour of financial practitioners and their effects on the market. A good review of the Behavioural Finance approach can be found in Shleifer (2000), Barberis and Thaler (2003), and Subrahmanyam (2007). There is a substantial amounts of work which has established the existence of several psychological underpinnings for FX market trad­ers’ behaviour, such as traders’ herding behaviour (Kim, et al., 2004), feedback trading (Aguirre & Said, 1999; Bj0nnes & Rime, 2005; Laopodis, 2005), heterogeneous expectations and beliefs (Frankel & Froot, 1990a, 1990b; Lto, 1990; MacDonald & Marsh, 1996; Oberlechner, 2001; Menkhoff, et al., 2009), traders’ overconfidence (Barber & Odean, 2000; Glaser & Weber, 2007; Oberlechner & Osler, 2008), and loss aversion bias (O’Connelle & Teo, 2009).

The second approach uses empirical micro­structure models to study the FX market micro­structure. A number of studies concerning FX markets have studied, through empirical models, the effect of order flow (Evans & Lyons, 2002a, 2004a; Berger, et al., 2006), market news arrival (Almeida, et al., 1998; Evans & Lyons, 2002b; Andersen & Bollerslev, 2003; Chang & Taylor, 2003; Evans & Lyons, 2004b), fundamentals trading on price movements and behaviour of market participants.

The Agent-Based Modelling (ABM) approach represents the features of a real financial market towards providing a behavioural explanation of the statistical properties of the financial market that are far from traditional assumptions of eco­nomics. The ABM approach models the financial markets as a group of interacting heterogeneous trading agents who can adapt to their market, learn from the information they obtain, and react dynamically to price movements and events on the market.

Although the existence of various approaches for studying the FX market, many properties con­cerning the FX market traders’ behaviour have not been revealed. We find that previous studies have acquired a large sample of low frequency dataset, or a small sample of high-frequency datasets, nei­ther of which are at account level. Based on that, we have studied FX market traders’ behaviour by combining a microscopic study of FX market trad­ers’ behaviour with agent-based modelling, which we have referred to as a bottom-up approach. The microscopic analysis of actual FX market traders’ behaviour aims to establish stylized facts of real FX market traders’ behaviour. To establish stylized facts, we use a large sample of high-frequency dataset of individual traders’ transactions at an account level. Using the identified stylized facts, we evaluate the collective behaviour of the agent­based models of traders.

Despite the existence of previous or contem­porary works on modelling the financial market using ABM, they have generally attempted to replicate some of the stylized facts of financial market. However, many of the studies pay less intention to identify the features and conditions that are responsible for the emergence of market behaviour. Some of the studies developed highly complex models of the financial market, which, due to their complexity, prevented them from identifying the conditions that are accountable for the reproduction of the real financial markets stylized facts. LeBaron (2001b) declares, it is important in agent based models not just to replicate features of real markets, but also to show which aspects of the model may have lead to them” (p. 226). In fact, it is important to have the minimal, simplified set of elements, that still ca­pable of resembling the market. Alfi et al. (2009a, 2009b) based their works on having the maximum mathematical simplification, which focuses on the essential elements that can generate the stylized facts of real market behaviour. Martinez-Jaramillo and Tsang (2009) identified the minimal set of conditions under which the statistical properties of price series in the artificial market resemble those of the real financial markets.

3.